Three different styles for bossing robots around.
“Prompt engineering” sounds fancy and scary, like something a programmer would do.
But it’s actually the process of telling a generative AI model what you want it to do.
This is something anyone can do, provided they can type, but doing it well can be more complicated.
Martin Waxman, adjunct professor at the York Schulich School of Business, recently shared with audiences at Ragan’s Employee Communications Conference the schools of thought on prompt engineering.
A conversational prompt is, well, just like the name. You tell the AI to do something, and it does its best to comply. For instance, you might ask it to draw a picture sitting on a park bench in winter. The AI will probably do an OK job delivering something that resembles your ask, but it won’t be particularly tailored to you or your needs because you haven’t provided any examples. It’s likely to return something serviceable but not exact.
The other option is what Waxman calls a structured prompt. As you might expect, this style of prompt requires a lot more work than conversational prompts.
A structured prompt should include these elements:
- Role, goal and audience: Who should the AI act as? Who is it creating content for? What is it trying to accomplish?
- Detailed, logical instructions.
- Expertise/experience: What does the AI know? What is it good at?
- Constraints/limitations: For example, edit, don’t write, ask questions, etc.
- Personalization to focus the AI on the audience and task at hand
- Provide examples.
- Specify output.
Obviously this will all be more work on the front end, but ideally, it should lead to a more refined product on the back end because the AI has a better view of your goals, the end product, what sort of “mindset” it should adopt and so on.
There is one final style of prompting that might strike you as a bit more off-the-wall. It’s called emotive prompting, and it relies on treating the AI as a person with emotions.
Yeah, it’s a little weird.
Basically, you’re playing off whatever emotions the AI might have, or might pretend to have, in order to get a better result. You might try offering the AI a bribe in return for good work. You might flatter the robot, saying it’s hard, but you know they can do it. You might even beg or imply your job depends on their work.
While it’s all a bit whacky, this is a brave new world of technology. Treating AI purely as a machine may not be the best approach.
Ultimately, being the most specific, thoughtful and careful is the way to go.
But maybe throw in a little flattery, too.
Watch Waxman’s full presentation below.
The post AI prompt engineering for comms pros appeared first on PR Daily.